Abstract

Due to the fact much of major depression research often focuses on methods of treatment or causal explanations emphasizing the role of psychological malfunctions, the responses of others to the depressed are largely overlooked. Although such responses are of little use to such explanations, the potential ability for depression to elicit positive changes in social relationships through signaling honest need has been suggested as way depression could be adaptive, making responses to it a critical piece of evidence. To examine if depression often comes with social benefits, we compared it to other proposed signals of need in a vignette study that also manipulated the amount and type of information about whether one’s need was honest and the presence of conflict within an imagined relationship. Overall, we found evidence for the predicted interaction between signal type and conflict on relationship outcomes as well as the predicted main effects of information and conflict. Contrary to our predictions, costlier signals did not result in more positive responses across signal types, with there also being little evidence in favor of the predicted signal by information interaction. However, depression outperformed all the other proposed signals of need on our four outcome measures across social circumstances, and our predicted relationships for the signals as a whole were largely supported when looking solely at the verbal request, crying, and depression manipulations. This suggests that depression is a relatively successful way of demonstrating honest need and eliciting support, consistent with signaling explanations.

Introduction

Long considered pathological sadness, depression has gained increased attention for its potential to function as a signal of need capable of bringing about positive changes in social relationships (Hagen, 2002, 2003). Referred to as the bargaining model of depression, this view shares much in common with signaling explanations for crying (Hasson, 2009; M. C. P. Hendriks et al., 2008), self-harm (Hagen, Watson, & Hammerstein, 2008), suicide attempts (Rosenthal, 1993; Syme, Garfield, & Hagen, 2016), and illness symptoms (Steinkopf, 2015; Tiokhin, 2016), all of which view the costs involved as a key component of eliciting important social support. Based mainly in signaling theory, such approaches have all benefited greatly from economics and game theory, with the result being explanations for the existence of such states that emphasize the causal role of one’s social environment rather than simply focusing on one’s internal state.

Due to the similarities between the ways in which they are suggested to achieve their hypothesized functions, there is much to be gained by comparing depression to the other proposed signals of need to better understand both the situations that bring them about and what determines differences in how they are responded to. In doing so, it should be possible to identify the fundamental similarities they share and create testable hypotheses that transcend signal type and, therefore, have the potential to further influence costly signaling theory.

Of all their similarities, the most basic is that positive responses are frequently expected so that the benefits received outweigh the costs involved on average. However, such responses are underexamined relative to the predictors of the symptoms involved and potential treatments, something which is especially true of depression (Billings & Moos, 1985) due to the suffering it causes to those who have it. The result of this is that one of the most important lines of evidence for an oftentimes adaptive explanation for depression and the other proposed signals of need is largely missing.

Thesis Outline

In this thesis, I briefly review the evidence for why the proposed signals of need are unlikely to solely be the result of malfunctions before reviewing the basics of signaling theory that tie them all together. I then review depression and how adaptive and nonadaptive approaches differ in their casual explanations. After examining the other proposed signals of need, I explore the empirical work that has been done on responses to the proposed signals before examining what should be expected from signaling theory. Lastly, I report the findings of a vignette study conducted with Edward Hagen examining if the proposed signals of need result in positive responses in the situations predicted by the previously mentioned signaling explanations.

Opposition to adaptive explanations and initial indications they are adaptive

In general, the dominant view of the proposed signals of need is that they are symptoms of harmful byproducts or malfunctions. Likely the main reason for this is the harm they cause the one experiencing them, which can be quite severe with the costlier signals (Kessler, 2012). However, subjective suffering and negative fitness outcomes are not inextricably linked. This is most obviously seen with negative emotions motivating fitness enhancing behavioral shifts, as is widely recognized to be the case with hunger, thirst, or pain (Nesse, 1990; Tooby & Cosmides, 2008). Just as these explanations attribute a functional role to negative experiences, the costs of proposed signals of need are suggested to be adaptive for their ability to increase the confidence of others who may be motivated to help that one’s need is genuine in ways non-costly displays could not (Hagen, 2003; Rosenthal, 1993; Syme et al., 2016).

However, to this point, the evidence suggesting that the proposed signals of need are adaptive comes primarily from outside indications of increased social support. For example, major depression (Bromet et al., 2011; Kessler et al., 2010), suicide attempts (CDC, 2014), placebo responses (Humphrey, 2002), and crying (van Hemert, van de Vijver, & Vingerhoets, 2011) are all more common than should be expected of maladaptive malfunctions due to the likelihood of strong selection pressures against those who incur extra costs in times of need. For the proposed signals of need to still be maladaptive with this prevalence, the traits responsible for such malfunctions would likely have to result primarily from environmental mismatches (Hidaka, 2012) or pleiotropic traits in which major benefits come from the genes responsible for the proposed signals of need (Williams, 1957). However, neither situation is safe to assume due to the proposed signals of need being found across a wide range of environments and social settings (Bromet et al., 2011; Syme et al., 2016) and the fact that genetic modularity often results in the potential of other genes to limit the negative effects of pleiotropy while preserving the benefits in polygenic traits (Hansen, 2003).

Another important indication that the proposed signals of need are often the result of the result of functional mechanisms is the fact they tend occur when individuals are in need and can benefit from others. This is seen in both depression and suicide attempts occurring most frequently in times of social adversity in which individuals are experiencing costs due to the actions of others or the loss of existing social relationships (Hammen, 2005; Mazure, 1998; Syme et al., 2016). Likewise, placebo responses, which have been proposed to be the downregulation of signaling through increased illness symptom severity (Steinkopf, 2015; Tiokhin, 2016), often happen in response to individuals receiving support after feeling a lack of it in conjunction with perceived need (Humphrey, 2002). Although not evidence of increased support, this connection between the proposed signals of need and times of need suggests that malfunction cannot be safely assumed as long as there are those around who would be motivated to help.

Social support in humans

Such a scenario in which costly emotional states lead to increased benefits is possible due to the degree to which humans cooperate with another, something which was necessary for our survival over much of our evolutionary history and is reflected in our adaptations today (Sugiyama & Sugiyama, 2003). Unlike most species, this support can often come from a wide range of individuals in humans, with kin beyond parents playing an important role in childcare across generations (Kaplan, Hooper, & Gurven, 2009; Meehan, 2005; Sear & Mace, 2008) and unrelated individuals often helping in a wide range of situations due to the benefits of both direct and indirect reciprocity as well as reputation gains (Alexander, 1987; Barclay, 2013; Jaeggi & Gurven, 2013). The result of these benefits is a situation in which many individuals may be potential targets of a costly signaling strategy.

While important at almost any time, cooperation is especially important in times of need. Illness, injury, and physical conflict with others have all been recurrent problems over the course of human evolution, with their outcomes often resulting in times in which individuals could not survive without short-term support (Sugiyama & Chacon, 2000). In these times, help often takes the form of temporary increases in provisioning or labor to decrease the workload of the ill (Sugiyama & Sugiyama, 2003). However, evidence of long-term support also exists with both Neandertals and modern humans being known to help those who are disabled (Trinkaus & Villotte, 2017). Beyond showing that such cooperation does not rely solely on expectations of direct reciprocity, the presence of long-term support for those who are unlikely to recover, together with the fact that humans can often reliably find support in times of need, suggests that lasting changes might be possible in times of need outside of instances of disability.

However, despite the likelihood of individuals benefiting from increasing their support of others in times of need and the widespread evidence that it often happens, devoting a large amount of resources to another is only likely to be worth it if the need is genuine. It is in these instances when the proposed signals of need are expected to be adaptive, as long as the result is an increase in the likelihood of support due to the signal enhancing one’s appearance of need and the benefits outweighing the costs of producing the signal (Zahavi, 1993). Although this limitation precludes the costlier signals from being adaptive in relatively minor instances of need, it does rule out adaptive explanations in times when one’s need is great and unlikely to be believed.

Costly Signaling

All organisms can benefit from altering the behavior, physiology, and psychological states of others, something which is seen from single celled organisms to social animals (Dawkins, 1982). Such changes can be accomplished through chemical or behavioral means, with information flowing from the signaler to the receiver in ways meant to benefit the one sending the message, regardless of whether or not it benefits the one receiving it (Dawkins, 1976). In social animals, members of one’s own species are often the target, with sexual selection and intrasexual competition creating many situations in which one’s reproductive success depends greatly on the responses of others (Clutton-Brock & Huchard, 2013; Daly & Wilson, 1983). The results of this are strong selection pressures favoring those who can make the most of these opportunities to outcompete others and receive the largest benefits through both cooperation and competition.

Of the strategies used to influence the behavior of conspecifics, costly signaling to provide believable information about one’s quality or current state is likely to be one of the most common and effective ways to do so (Zahavi, 1975). A signal differs from a cue in that it is intentionally sent by an organism to alter the behavior or thoughts of other individuals, while a cue is any piece of information an organism uses to alter its behavior or physiological state that was created without the intent to do so (Maynard Smith & Harper, 1995). Since signals are created for the purpose of communication, they are by definition easy to fake, especially in situations where they can be disconnected from the qualities being signaled (Zahavi, 1975). This coupled with the selection pressures resulting from the fact that conflicts of interest are ubiquitous among living things (Trivers, 1974; Zahavi, 1993) means that there is much to gain from successfully faking a signal as one could receive the benefits without paying the costs associated with the underlying quality. For this reason, strong selection pressures exist favoring adaptations to spot fake signals, something which has been shown to exist in many species (Hasson, 1994).

To overcome such counter-adaptations to cheating, an organism may produce a signal that cannot be achieved without meaningful costs in order to increase the believability of the signal (Zahavi, 1975). The result of such signals are relatively honest displays compared to those which involve lesser costs so long as the costs involved with the signal make the signal only worth producing for those who have the underlying qualities associated with the signal (Zahavi, 1993). Although originally formulated as an explanation for the role of costly traits in attracting mates (i.e. colorful plumage or large physical weaponry) (Gangestad & Scheyd, 2005; Zahavi, 1975), there is no reason why it cannot be applied to the proposed signals of need as long as their costs are prohibitively high for those who are not in need.

Depression

Of all the proposed signals of need, major depression has received the most attention and likely causes the most suffering. Although often overlooked due to its indirect connection with mortality (Patel, Araya, & Bolton, 2004), it is the leading cause of nonfatal disease burden in the world and the second greatest contributor to global years lost to disability (Ferrari et al., 2013; Slavich & Irwin, 2014). Around 8% of people in the U.S. experience major depression each year, and almost 20% will experience major depression in their lifetime (Kessler et al., 2010). In general, similar prevalences are often reported for depression as a whole, with there being some evidence that it is more often diagnosed in developed countries (Bromet et al., 2011). However, in no country is there a good understanding about how people respond to the depressed.

Major depression typically involves a large-scale reduction in activity along with persistent negative feelings (Fried & Nesse, 2014). These symptoms take many forms but generally fall into the categories of insomnia, weight loss, fatigue, agitation, attention problems and feelings of worthlessness. As these symptoms can accompany normal sadness, length is primarily used to distinguish it from minor depression, with DSM-5 requiring the existence of at least five symptoms for at least two weeks (Association, 2013). However, it often also differs in severity, as is seen with depression’s connection to increased suicidal ideation (Bostwick & Pankratz, 2000). From an evolutionary perspective, it can be described as partial to complete abandonment of normally fitness enhancing behavior that also results in a reduction of benefits one provides to others (Hagen, 2003).

In addition to these symptoms used by DSM-5, depression is often associated with impaired social functioning (Hirschfeld et al., 2000). Although causality is likely to go both ways, such impairments encompass almost the totality of social functioning with depressed individuals having smaller social groups, being more likely to perceive social interactions as negative, and having lower levels of social support (Gotlib & Lee, 1989). Beyond simply being a symptom of depression, a lack of improvement in various measures of social functioning is also associated with reduced chances of remission (Billings & Moos, 1985), although even those who return to a non-depressed state may still have lower measures of social functioning than healthy controls (Weissman & Paykel, 1974).

Regardless of the exact combination of symptoms, most cases of depression are closely tied to major negative life events (Hammen, 2005). This is clearly seen when comparing those with depression other groups, with depressed individuals often reporting at least twice as many instances of negative events than nondepressed individuals (Mazure, 1998) and more negative events than those mental disorders like schizophrenia and bipolar disorder across a wide range of studies (Paykel, 1994). In addition to differences in recent experience, the connection between negative events and depression is also seen in the high percentage of depression cases that are preceded by negative events and the rapid onset of depression following the event, both of which are seen in a largescale study of female twins that found 80% of depression cases were preceded by at least one negative life event and symptoms tending to occur within one month of the event (Kendler, Karkowski, & Prescott, 1999).

This strong connection between negative life events and depression along with the need and distress that likely accompanies these events has led to the suggestion that this relationship is causal (Kendler et al., 1999). Support for this comes from the fact that the relationship between negative events and depression holds true when looking solely at events outside of one’s control, which shows that the connection is unlikely to be driven solely by depressed individuals creating situations where negative events are likely to be common (Hammen, 2005; Kendler et al., 1999). In addition to this, twin studies have shown that one’s history of negative events is still a strong predictor when controlling for genetic similarity and that part of the heritability of depression stems from the heritability of negative events, like divorce and family conflict (Kendler & Baker, 2007; Kendler et al., 1999).

Of the many types of negative life events associated with depression, those which involve social adversity tend to be more strongly associated with depression (Hammen, 2005; Kendler et al., 1999; Mazure, 1998). Like the relationship between negative events as a whole and depression, this relationship is also seen cross-culturally, with the loss of a loved one, threat of divorce, or abusive relationships all increasing one’s chances of becoming depressed (Kendler et al., 1999, 1995). Of these predictors, physical assault has been found to be the strongest predictor of depression in the few studies in which it has been examined (Kendler et al., 1999, 1995) and may have been much more common throughout our evolutionary history than is often thought based on rates of spousal abuse in small-scale societies (Stieglitz, Kaplan, Gurven, Winking, & Tayo, 2011).

Except for the loss of a loved one, conflict is likely to often be present in all the previously mentioned forms of social adversity and may be one of the most meaningful aspects in bringing about depression. For example, breakups in both romantic and cooperative relationships are often preceded by conflict (Amato & Hohmann-Marriott, 2007), and surely contain it when one side wants reconciliation and the other does not. Likewise, abusive relationships are bound to be conflictual, with one side holding the power to harm the other and doing so. When looking solely at conflict as a predictor, it is often reported to be linked with depression (Asgeirsdottir, Sigfusdottir, Gudjonsson, & Sigurdsson, 2011), with both conflict without abuse and conflict among family members also being strong predictors of depression in Western societies (Kendler et al., 1995). Although too few studies exist outside of Western societies to be sure of the generalizability of this, initial studies are consistent with this, with conflict outside of the family being a predictor of depression among the Tsimané (Stieglitz et al., 2015).

Nonadaptive explanations

Explanations for the relationship between social adversity and depression often involve the malfunction of mechanisms relating to low mood, which is almost universally considered adaptive when it functions properly (Hagen, 2003; Nesse, 1999; Wakefield & First, 2012). One influential version of this is the stress-diathesis model , which suggests that stressful situations interact with one’s predisposition to depression to bring about depressive symptoms (Monroe & Simons, 1991). In general, such approaches do not make distinctions between different types of stress and their effects on risk of depression nor do they predict different outcomes based on variables outside of stress levels and personality differences. From this perspective, the social adversity often associated with depression may simply be one way for extreme levels of stress to accumulate thereby turning what would normally be short-term sadness into long-term depression.

Such approaches also generally do not make predictions on the types of responses the depressed are likely to receive, although there are exceptions. This is due in large part to the previously mentioned costs involved, which shift the focus of most research towards treatment (Patel et al., 2004) and increases the size of the benefits needed for it to adaptive. When predictions are made about responses, they tend to involve depression resulting in negative responses due to strain the depressed’s anhedonia and impaired social functioning puts on relationships (Coyne, 1976). From these perspectives, the costs of depression are simply too great to warrant a focus on the positive responses or too unexpected to be examined.

The bargaining model of depression

Unlike these approaches, the bargaining model of depression emphasizes both the roles of social adversity and potential benefits in its explanation for why most instances of depression happen (Hagen, 2002 , 2003). It views depression as an oftentimes adaptive response to social adversity or other hardships in which support from others or more beneficial cooperative relationships can be gained. Since any biological mechanism can fail, maladaptive cases of depression are still expected to exist under this model (Hagen, 2011). However, in many instances, it is expected to serve as a viable option to elicit more social support for those who may have few other options.

The bargaining model of depression is based on the idea that for much of our evolutionary history the small groups humans lived in created many situations in which certain forms of cooperation were unlikely to be easily replaced by others while also allowing for harmful or unfair social relationships to develop (Hagen, 2003). In these instances, one’s monopoly on one or more types of cooperation could be leveraged to change how one is treated if others value the depressed’s cooperation enough and view the depressed as being in honest need. In other words, depression could function similarly to a labor strike, demonstrating that one’s need was genuine; and that if the target of the signal wanted to continue receiving benefits from the depressed, he or she would need to increase support towards the signaler.

Like with the other proposed signals, the costs experienced by the depressed play an important role in its function as an indicator of honest need (Hagen, 2003). As previously mentioned, depression often involves substantial psychological pain and a largescale reduction in one’s normal routine of fitness enhancing behavior. This means that depression is likely unprofitable to fake as the costs involved are likely only worth experiencing by those who have high levels of need, thereby increasing the confidence of others that the signal is honest as is suggested with the other costly signaling hypotheses. Importantly, these costs only need to be maintained until the signaling function is thought to be met, with symptoms being expected to subside shortly after indications that future changes in social support are likely to come. This increases the chance that the benefits received will outweigh the costs relative to explanations that require continuous signaling and results in clear predictions about why depression ends when it does.

Despite predicting many cases of depression to stem from situations in which it was adaptive ancestrally, the bargaining model does not suggest that depression should go untreated. This is because adaptation is not to be confused with subjective wellbeing, which is a much more important concern in every area except for explanations of why depression exists (Cosmides & Tooby, 1999). However, it does suggest an explanation for why most medical treatments perform similarly to cognitive therapy (Hollon et al., 2005) and placebos (Walsh, Seidman, Sysko, & Gould, 2002) in that all three methods demonstrate the existence of a knowledgeable medical professional giving active support to the depressed. In other words, such treatments may provide sufficient cues that the signal has worked, even though none may treat the underlying cause.

Other proposed costly signals of need

As previously mentioned, the bargaining model of depression is not unique as an oftentimes adaptive explanation for harmful symptoms, with crying (Hasson, 1994; Hendriks et al., 2008), self-harm (Hagen et al., 2008), suicide attempts (Rosenthal, 1993; Syme et al., 2016), and illness symptoms (Steinkopf, 2015; Tiokhin, 2016) having all been argued to signal need.

Of these proposed signals, self-harm and suicide have the most in common with depression. Both are much more common for their costs than should be expected for malfunctions (Hagen et al., 2008), with suicide attempts being found across a wide range of cultures (Syme et al., 2016; Weissman et al., 1999). As with depression, self-harm and suicide attempts tend to come in times of social adversity like divorce or neglect, both of which are situations when it is likely that social support is both greatly needed and difficult to obtain (Hagen et al., 2008). This connection to social adversity is clearly seen among instances of suicide attempts in the HRAF database, in which 77.4% of attempts were found to be linked to times of social conflict, and 88.7% were linked to fitness threats in general (Syme et al., 2016). Importantly, 84.9% of attempts were linked to situations in which attempters were powerless to resolve their situation alone (Syme et al., 2016), further indicating its potential as a signal of need in times when other less costly methods of bargaining are less likely to be successful, assuming the benefits are large enough on average.

Unlike depression, where the costs are directly felt by the signaler and his or her targets, the bulk of the costs associated with suicide attempts are only felt in times that result in death or severe bodily harm. For this reason, the way others view and respond to suicide attempts is likely to be strongly influenced by the probability that the attempt would have been successful (Rosenthal, 1993). Differences in these probabilities result primarily from the method of the attempt, which can range from little to no risk to almost certain death (Weiss, 1957). Of the possible risk values on this continuum, the best candidates for bargaining are those in which there is a “non-negligible” but relatively unlikely chance of death (Rosenthal, 1993) as those which involve minimal risk are likely to be thought of as manipulative and those which are riskier than need be not being what is expected of fine-tuned adaptations. Therefore, a key prediction of the bargaining model of suicide is that attempts should often result in the attempter surviving (Syme et al., 2016). This is clearly the cases across the world, with attempts that do not result in death far outnumbering those that do (CDC, 2014) and individuals often choosing less risky forms of suicide when methods with higher probabilities exist (i.e. overdosing on pills instead of jumping off of a tall building) (Rosenthal, 1993).

Of the proposed signals, crying has the longest history of being examined as a signal which often results in benefits and is the least controversial due to its low costs (Hendriks et al., 2008). For example, infant crying is almost always considered adaptive, although how it functions to elicit support is still debated, with many proposed explanations involving functions outside of signaling honest need like demonstrating quality, blackmailing parents through increased risk of predation, or manipulating parents into investing less in other offspring (Lummaa, 1998). However, like the other proposed signals of need, it is more common when one is in need and can benefit from the support of others. This is seen both observationally, with crying bouts being shorter in cultures which have higher levels of parent-offspring contact (Lummaa, 1998) and experimentally, with longer crying bouts being associated with shorter holding times (Hunziker & Barr, 1986).

Like infant crying, adult crying is a human universal, with emotional tears likely being unique to humans (M. C. Hendriks et al., 2008). Due to differences in what motivates support as well as differences in both the prevalence and the costs involved, it is likely that adult crying often differs from infant crying in both function and the many of the underlying psychological mechanisms. However, in both infants and adults, crying is an effective signal that something is wrong, with tear production and the facial sadness that often accompanies crying both increasing the perception of sadness by others (Hasson, 2009) and humans being well equipped to distinguish tears of joy from tears of sadness (Hendriks & Vingerhoets, 2006).

Unlike the other proposed signals of need, adult crying is frequently reported to occur without a specific external referent, with even the crier often left to wonder why it happened when it did (Vingerhoets, Cornelius, Van Heck, & Becht, 2000). However, when looking solely at bouts of crying that can be traced to a cause, it is clear that social factors are the strongest predictors, especially when conflict or loss is involved (Vingerhoets et al., 2000; Young, 1937). In addition to showing that crying appears in the right contexts for it to potentially be adaptive, the fact that tears which can be traced to a cause rarely occur outside of times of social need hints that it is often an honest signal. In line with this reasoning, the absence of an effective sadness display has been shown to reduce supportive behavior, suggesting that properly displaying sadness (with which crying helps) increases the confidence of others that one is in need (Barr-Zisowitz, 2000; Diminich, 2016). However, the fact that adults often cry out of sight of others suggests that its occurrence cannot be explained solely by a signaling explanation (Vingerhoets et al., 2000).

Compared to depression, self-harm, and suicide attempts, the costs of crying are much less severe, and it is unclear if its costs are significant enough to result in honest displays on their own (Hasson, 2009). One important area in which the costs of crying are likely to be felt is through reputation hits, with crying often lowering how others view a person (Hendriks & Vingerhoets, 2006). While true of both sexes, this may be particularly true if the target of the signal is a man, as men have been shown to simply avoid crying people more often than women in experimental settings. Although not directly linked to crying, sadness, which almost certainly accompanies most instances of crying, tends to slow both cognitive and motor functioning (Diminich, 2016; Ekman, 2003; Ellgring & Scherer, 1996) thereby imposing at least some opportunity costs surrounding and during bouts of crying, something that may be directly seen with the blurred vision that comes with tears (Hasson, 2009). When sadness is already present, crying may simply exacerbate this distress, as is seen physiologically with the increased hearts rate and blood pressure that typically accompanies crying taking time to return to normal levels after crying (Hendriks et al., 2008)

Although tied to a different type of need and not included in the study presented here, illness symptoms provide another strong candidate as a way for costly symptoms to function as a signal of need and are, therefore, worthy of being included in this comparison. Referred to as the signaling theory of symptoms, the idea is that while illness symptoms initially arose as the result of defenses against disease or a pathogen’s circumvention of these defenses, they quickly became cues for conspecifics to alter their behavior to better avoid illness, take advantage of a rival’s weakened state, or help kin and close allies (Steinkopf, 2015; Tiokhin, 2016). Although the first two conditions would likely favor the suppression of symptoms, times in which help could be received may have favored times in which symptoms would become signals of honest need as opposed to just cues. The reason for this is once strategies for providing support to the sick became common enough, selection would likely favor those who could fake symptoms, which would then favor anti-cheating adaptations that require symptoms that were hard to fake to be believed. In these instances, those who upregulated already present symptoms beyond the optimum for disease fighting would provide a more honest, costly signal and, therefore, be more likely to satisfy the anti-cheating adaptations of others and increase their chances of receiving aid.

Formulated as an explanation for placebo responses, the best evidence in favor of the signaling theory of symptoms comes from the observation that placebo responses tend to occur after signs that care has been given, with the placebo response simply being viewed as the return to symptom levels appropriate for disease fighting without the added signaling component (Steinkopf, 2015). Therefore, like the other proposed signals of need, its occurrence is expected to be highly dependent on social context in both predicting its occurrence and its cessation. This is seen experimentally in that placebo responses which often work in the presence of healthcare providers can be completely negated if the person is instructed to complete their treatment alone (Hashish, Ho Kee Hai, Harvey, Feinmann, & Harris, 1988). However, presence alone is also not enough, as should be expected of a signaling explanation. Instead, demeanor is also highly important, with facial expression indicating care and concern resulting in more placebo responses than neutral or disinterested faces (Benedetti, 2009; Steinkopf, 2015).

Key similarities shared by the proposed costly signals of need

Viewed collectively, the proposed signals of need clearly share some important similarities which can allow for generalizations about when they should be most common and under which situations they are most likely to be adaptive. When it comes to prevalence, their hypothesized functions as signals suggests they should all be more common when there is a high degree of private information held by the signaler (i.e. when the targets of the signal do not have information about the honesty of need) (Hagen, 2003). This is due to the previously mentioned ubiquity of conflicts of interest, which incentivizes deception and favors strategies in which individuals are only likely to increase their support of others if their counter-cheating adaptations are satisfied. For this reason, signals should be more common when the target has little information to confirm the signaler is in need and less common when such information exists as the signaling component of any trait or state is created primarily to provide another organism with information. Indeed, in instances of public information, the expectation is that both parties should be able to bargain quickly as there is nothing in the way of both sides figuring out how the benefits should be divided (Kennan & Wilson, 1993).

One basic similarity that allows for more detailed predictions is that all the proposed signals are predicted to result in benefits that make up for the costs involved, at least on average. Although this seems more relevant to what happens after the signal, it also has important implications for when signals might be employed since smaller forms of support are likely to be sufficient for the less-costly signals to be worth it, with the costlier signals requiring larger gains all else being equal. Therefore, it is reasonable to expect the less costly signals to be more common, something which is seen with instances of crying far outnumbering cases of depression, which far outnumbers the cases of suicide attempts (Kessler & Bromet, 2013; Nock & Kessler, 2006; van Hemert et al., 2011).

One way to reduce the average costs of the proposed signals, and therefore increase the times which they may be applicable, is to only engage in them in situations where large payoffs are relatively likely (Rosenthal, 1993). Although little has been done to investigate the existence of such mechanisms, it is a reasonable trait to expect of the proposed costly signals of need if they are adaptive. This is because the absence of such systems would decrease the amount of ways the proposed signals could have evolved as adaptations due to the role selection would play in disfavoring strategies that led to dangerous misfires and worsened conditions in times in which such outcomes may be the costliest. Without these mechanisms, the adaptive nature of the proposed signals of need may be confined only to times of extreme need, as such situations may reduce the costs of signaling to such an extent that the relevance of the likelihood of support to levels where it becomes much less relevant (Hagen et al., 2008; Tiokhin, 2016).

Evidence for this may be seen with findings that suggest that some of the proposed costly signals of need are more common in situations where they are more culturally acceptable and where help may be more available. For example, the size of the female biased sex difference in depression was found to differ in relation to gender equity, with more equitable societies experiencing larger gender gaps despite overall lower rates of depression (Hopcroft & Bradley, 2007). This same pattern is seen with crying, with women crying more often in countries with greater gender equality, something which clearly runs counter to models which emphasize the causal role of stress overloads (van Hemert et al., 2011). One possible reason for these relationships is that women in more equal societies signal more because they are more likely to receive support. Consistent with this interpretation is the fact that while crying tends to not result in improved emotional states in general (Hendriks et al., 2008), it does result in women feeling better after crying in countries with relative equality (Becht & Vingerhoets, 2002) and when social support is present (Rottenberg, Bylsma, & Vingerhoets, 2008).

In addition to not engaging in a signaling strategy that is likely to result in insufficient support, only investing in a signal of need when one does not have any less costly alternative options is another likely way to reduce the average costs of the proposed signals. This prediction is clearly built into the unified model of anger and depression, which proposes that anger and depression may be complementary strategies, with aggression being favored by those who have the formidability to bargain effectively through the use of force and depression favored by those that cannot (Hagen & Rosenström, 2016). Support for this comes from the finding that differences in upper body strength account for much of the female biased sex difference seen in depression. Importantly, this sexually dimorphic pattern of women experiencing depression symptoms more often than men is also seen in all the other proposed costly signals examined here (CDC, 2014; van Hemert et al., 2011), suggesting that less costly bargaining methods are reliably favored, assuming there is a similar scarcity of alternatives for women relative to men.

Although not derived from the similarities of the theories explored here in the same way as the previous traits, a key similarity shared by the signals is that the costs associated with signaling will almost always be less than the costs of the symptoms themselves since some level of symptomology is likely to be present that is not tied to a signaling function. In other words, the cost of engaging in a costly signal of need is often going to be less than the symptoms the recipient sees. This is clearly seen with illness symptoms if the signaling theory of symptoms is correct, in that much of one’s symptoms will be the result the body’s defense mechanisms or a pathogens circumvention of these defenses even in instances in which they are upregulated for a signaling function (Steinkopf, 2015; Tiokhin, 2016). Therefore, the costs of signaling may only be a small part of the totality of the symptom. Although such a situation is harder to imagine with depression or suicide attempts, extreme cases of conflict or abuse may result in situations in which the difference between a depressed and a nondepressed state may be much closer than normal (Hagen, 2003).

Benefits

For the proposed signals of need to be worth the costs, substantial benefits in terms of social support are expected by their corresponding theories, either in the form of support or renegotiated social relationships. When it comes to depression, self-harm, and suicide, the most likely benefits are expected to be long-term changes in how one is treated (Hagen, 2003; Hagen et al., 2008; Syme et al., 2016). This can take the form of reconciliations in failing relationships, increased fairness in cooperative relationships, or an increase of resources being directed towards an individual. Of all the benefits, a reduction in physical abuse may be one of the most important, with physical abuse likely being a common occurrence over our evolutionary history based on the high prevalence of it in traditional societies (Stieglitz et al., 2011).

The benefits of the less costly signals are likely to be more varied due to their low costs allowing smaller benefits to make them worthwhile. These benefits may therefore be more short-term in nature and come from relationships without large conflicts of interest. In regard to crying, a demonstration of need beyond a verbal request may increase the chances of short-term changes in resource allocation or interventions into social conflicts outside of the signaler-target relationship. When it comes to illness symptoms, greater symptomology is expected to increase the likelihood of support when one cannot satisfy their own needs (Steinkopf, 2015; Tiokhin, 2016) and may also decrease the risk of reputation hits that might come from the belief that one is faking illness , both of which might be well worth a minor increase in symptomology which might already exist.

Whether long-term or short-term, social support is often classified as emotional or instrumental (material support), with there currently being little empirical evidence on their relative importance (Jacobson, 1986). However, if emotions are viewed as motivators to guide behavior towards adaptive outcomes (Nesse, 1990; Tooby & Cosmides, 2008), it is unlikely emotional support is the ultimate goal. Instead, emotional support may be the desired outcome of proximate mechanisms which favor greater investment in relationships to increase the likelihood of future instrumental support if one needs it (the ultimate benefit). It is also conceivable that emotional support might track achieved support so that emotional support is high when one’s needs are being met. If this is the case, changes in emotional support may still be informative to signaling approaches as achieved support is likely to cause individuals to believe there is less need to signal, assuming an honest display needs to only be made once.

Although less meaningful when examining responses to the proposed signals, the distinction between received and perceived social support is likely more useful when considering how information about social support is stored, with received support being material support and perceived support being one’s perceptions about the likelihood and adequacy of future support (Norris & Kaniasty, 1996; Wethington & Kessler, 1986). Unlike the relative importance of material and emotional support where the evidence is ambiguous, here, the evidence suggests that perceived support is the more powerful of the two in predicting positive outcomes, with the two not being as closely related to each other as often expected (Haber, Cohen, Lucas, & Baltes, 2007; Norris & Kaniasty, 1996). Although evident in many areas, the disconnect between received and perceived social support is particularly noticeable in the finding that perceived social support often decreases in times of need, even in situations in which one is likely experiencing more material support than in one’s life before the need (Norris & Kaniasty, 1996).

When examining the costly signals of need, the gap between current levels of received social support and the potential support that could be gained through a signaling strategy is likely to be the most important measure as this is what defines the size of the benefits. As this would involve information about both one’s level of need as well as information about the likelihood and size of potential help, it may essentially be a measure of perceived support as defined above. Therefore, perceived social support might form an important link between the severity of one’s situation and the amount of help deemed sufficient to avoid or stop a signaling strategy. In connecting the two, this conceptualization has the potential to explain why low levels of perceived social support are a strong predictor of depression (Kendler & Gardner, 2014; Leskelä et al., 2008; Park et al., 2016) and a potential predictor of the symptom upregulation found in placebo responses (Humphrey, 2002).

Responses to the proposed costly signals of need

Most studies examining responses to depression have been observational and focused on responses outside of changes in social support. In general, most report aversive responses and high levels of rejection towards the depressed, and depressed individuals are often linked to impaired social functioning (Evraire & Dozois, 2011; Gotlib & Lee, 1989; Hirschfeld et al., 2000). One explanation for this is that depressed individuals engage in behaviors that drive others away. In line with this reasoning, both excessive reassurance seeking and negative feedback seeking, which often accompany depression, have been shown to elicit negative emotional responses from others and increase the chance of rejection from others (Joiner & Metalsky, 1995; Starr & Davila, 2008). Negative interactions are also seen in families as depressed children tend to report less support and more negative responses from parents than non-depressed children, although it has not been demonstrated that such interactions did not lead to depression in the first place (Messer & Gross, 1995; Schwartz et al., 2012).

Beneficial responses to the depressed are also reported. This is seen in that depression has been shown to reduce the likelihood of aggression within families, particularly in families already experiencing high levels of conflict (Sheeber, Hops, & Davis, 2001). There is also limited evidence that depression can result in increased levels of support from others outside the family. For example, non-depressed individuals were shown to be more cooperative and exhibit more caretaking behavior towards their college depressed roommates over the course of their time living together (Hokanson, Loewenstein, Hedeen, & Howes, 1986). Likewise, a vignette study found that individuals were more likely to provide advice and support to depressed individuals than anxious individuals despite similar levels of distress (Stephens, Hokanson, & Welker, 1987).

Responses to crying have received more attention, often from a signaling perspective. These responses are generally positive when it comes to infant crying and instrumental outcomes like the amount of care directed towards the infant or its proximity to its parents (Lin & McFatter, 2012). However, crying is also associated with increased distress and other negative emotions in parents and may increase the child’s risk of maltreatment if it is in poor condition (Lin & McFatter, 2012; Soltis, 2004). Responses to adult crying tend to be similar with negative responses typically involving emotions and positive ones involving the likelihood of providing support. For example, individuals faced with crying scenarios in a vignette study experienced more negative emotions relative to controls while also being more likely to provide emotional help and hide negative feelings (Hendriks et al., 2008). Likewise, another experimental study found that while crying increased social support, it came at the cost of being perceived as helpless (Vingerhoets, van de Ven, & van der Velden, 2016). It is important to note, however, that this trend is not always the case as crying can elicit positive emotions like perceived friendliness (Vingerhoets et al., 2016) and result in negative outward responses like confrontations or mocking at the extreme end (Wagner, Hexel, Bauer, & Kropiunigg, 1997).

Unfortunately, there are few studies examining how individuals respond to suicide attempt survivors. In those which exist, stigmatization is commonly reported within and outside families (Frey et al., 2017; Scocco, Castriotta, Toffol, & Preti, 2012), with perceptions of survivors as being weak, selfish, mentally ill, and antisocial all being commonly reported (Batterham, Calear, & Christensen, 2013; Frey et al., 2016a; Tzeng & Lipson, 2004). Although these types of stigmatization have been reported across relationships types, the likelihood of stigma has been found to differ based on who the survivor is to the person being questioned. This is seen in a study which found that individuals felt the most stigma from members of their preexisting social networks than mental health providers, with family members providing more stigmatization (57.1% of the time) than romantic partners (41.2% of the time) and close friends (28.2% of the time) (Frey et al., 2016b).

A factor analysis of responses to suicide survivors found that stigma related to suicide attempts could largely be described by three factors: weak, crazy, and distressed (Corrigan, Sheehan, & Al-Khouja, 2017). When it existed, prejudice was described by two factors (fear/distrust and anger) and discrimination by three (avoidance, disdain, and coercion). A later vignette study which examined these factors across mental health problems found that participants had higher levels of stigmatization towards imagined characters who attempted suicide (both survivors and non-survivors) than depressed or mentally ill characters (Sheehan, Dubke, & Corrigan, 2017). In addition to this, the suicide vignettes also resulted in more anger than the other conditions; however, discriminatory behavior was about equal regardless of whether the attempt resulted in death or accompanied depression.

Although negative responses to suicide survivors are more often studied and reported, positive responses have also been found. Increased social support and changes to important relationships are often reported to follow suicide attempts with some indication that these effects may hold long term (Stengel, 1956). For example, a study of 100 women who survived suicide attempts found that in 75 cases, individuals gained identifiable benefits through the attempt, with 41 individuals benefiting from reconciliations with others (Lukianowicz, 1971). As for the type of support received, emotional support seems to be more commonly provided than instrumental support based on initial evidence (Frey et al., 2017); however, it is currently unclear how this might change with social context.

Unfortunately, the evidence regarding responses to suicide in traditional societies is indirect or anecdotal. Unlike Western countries, where suicide is often viewed as pathological (Hidaka, 2012), members of traditional societies have been reported to view suicide attempts as cries for help rather than mental illness (Shostak, 1981), suggesting that help is at least occasionally provided or reliably expected. Further evidence for this comes primarily from the situations in which suicide attempts are made, with suicide attempts often being described as ways of escaping unwanted marriage arrangements, persistent abuse, or a lack of support in obtaining mates (Gutiérrez de Pineda & Muirden, 1948; Hilger, 1957; Karsten, 1935; Tessmann, 1930; Wilson, 1960). Although most reports focus mainly on suicide attempts that result in death, there is some indication that the threat of it or a survived attempt can result in increased support, as was the case with the threat of suicide helping a !Kung woman escape an arranged marriage (Shostak, 1981).

Why responses differ

Compared to the empirical examinations of the responses of others to the proposed signals of need, the theory is much more advanced and responsible for several testable hypotheses. Likely the most basic is that costlier signals will elicit greater benefits all else being equal. This is mainly due to the previously mentioned relationship between the costs of a signal and its honesty, which results in costlier signals better satisfying the anti-cheating adaptations of others than less costly alternatives (Zahavi, 1993). In addition to this, the requirement that the benefits received need to outweigh the costs of producing the signals on average also suggests that the costlier signals will often result in more substantial support as increased costs must be met by increased benefits for them to be adaptive.

One area where all else is not equal and less costly signals may outperform costlier ones is times of minor need. Though not yet demonstrated empirically, it is reasonable to suspect an interaction between signal strength and need whereby stronger signals only tend to result in more positive responses in times where the signal is proportional to the need. In other words, too strong of a signal in times of minor need might send a stronger message of dysfunction than need. This may be problematic for the signaler because although dysfunction might still mean support is needed, its presence likely indicates that there is less to gain through supporting the individual through direct reciprocity regardless of relationship type and inclusive fitness if the supporter is related to the signaler and the perception of dysfunction is substantial (Sugiyama & Sugiyama, 2003).

Although less examined empirically than the relationship between the cost of a signal and its honesty, it is also reasonable to expect that existing conflict will require costlier signals to be effective. This is mainly due to the likelihood that existing social conflict means larger conflicts of interest are involved, which would therefore require the signaler to be in relatively greater need to be worth helping than in instances with no conflict. High degrees of conflict with the signaler may also increase one’s scrutiny of the signals, which would once again favor costly displays. In addition to fitting signaling theory, such a prediction is also consistent with the previously mentioned predictive relationship between conflict and depression, as depression is an extremely costly state to experience. Since the proposed signals of need often come in response to social adversity and are predicted to serve as a way to increase social support, it is also reasonable to expect that relational level differences are likely be more important than individual level differences in affecting how others respond to signaling. This seems to be borne indirectly out in the few studies that exist on the topic with relational level differences being the better predictor of the amount of disclosure suicide survivors were likely to give family members (Frey & Fulginiti, 2017; Fulginiti, Pahwa, Frey, Rice, & Brekke, 2016), something that is associated with more positive responses from family members and lower depression scores (Frey et al., 2016b). However, it is unclear if it was disclosure itself or being in a more open family that resulted in positive responses.

Another situation when relational level variables might be important is in times of divergent welfare tradeoff ratios (i.e. when one partner values the other more than they are valued in return) in relationships defined by reciprocal altruism (Sell, 2005). In these instances, the less valued partner may have more to gain in supporting the more valued individual due to the benefits of maintaining future support, assuming the reason for the differential WTR is not due to kinship (i.e. younger kin valuing older kin less) or disability (which would make reciprocal altruism unlikely). In addition to this, such times of need would also create opportunities for the less valued partner to close the gap in WTR for future benefits by providing important support, potentially making support more likely than if the roles were reversed.

It is important to note that the relative importance of relational level differences does not mean individual differences are likely to be unimportant. This is especially the case when these differences contribute to differences in one social environment (Kendler et al., 1999). Although not evidence of responses to need, such a situation is seen with many of the social situations associated with depression being heritable like divorce and family conflict (Kendler & Baker, 2007). If positive relationship outcomes are also heritable in a similar way as is reasonable to expect, then individual level differences would also affect how much help is provided in ways that would be hidden by relational variables, rather than simply leading to differences in likelihoods of occurrence.

Beyond one’s social environment, individual differences may also result in differences in how effective various signals and their alternatives are, thereby influencing how one is responded to. Although not a costly signal as envisioned here, anger is clearly involved with signaling displays and should be expected to work better for those with more formidability (Hagen & Rosenström, 2016). As previously mentioned, this has been suggested to explain the sex difference in depression, with women being less likely to effectively bargain through aggression. However, if depression and anger from complementary bargaining strategies, it is also conceivable that men may have more to lose from depression. Although speculative with depression, this appears to be the case with crying, with men often being responded to less favorably to than women (Cretser, Lombardo, Lombardo, & Mathis, 1982; Jesser, 1989) although this is not a universal finding (Hendriks et al., 2008).

Study Aims and Predictions

The relative lack of studies on responses to the proposed signals of need thus far has meant that a key line of evidence regarding the plausibility of their adaptive nature is underdeveloped, something which is compounded by the fact that many of the studies examining the costlier signals are focused solely on negative responses. In this study, we examine responses to many of the proposed signals of need in response to different social circumstances (i.e. high or low conflict and the amount and direction of information about honest need). Although this is done for all the signals in the confirmatory portion of the study, we turn our focus to depression and the less costly signals in our exploratory analyses.

In response to the previously mentioned hypotheses, we predict that stronger signals will result in higher levels of both emotional and instrumental social support (i.e. likelihood of providing help or how much money one is willing to give) towards the signaler. Specifically, we predict that the signals will have the following rank order form low to high: no signal (control), verbal request, facial sadness with crying, depression, depression with display of suicidal ideation, and suicide attempt. We also predict that anger will have a positive impact on helping with a similar magnitude to the depression condition. Unlike the hypothesized signals of need, we predict that the schizophrenic break condition will decrease the likelihood of help relative to the control since it indicates cognitive dysfunction and its symptoms are not expected to have a signaling component.

Although instances of private information are likely to result in the highest prevalence of signaling behavior, it is unlikely that they will result in the most helping behavior. Instead, we predict that instances of public information of honest need are likely to result in the most help as external information of need is likely a more reliable indicator of need than signals under most circumstances. In line with this reasoning, we also predict that private information will result in more support than public information of cheating as even costly signaling attempts may not overcome reliable information of a lack of need. Due to the previously mentioned likelihood that conflict increases skepticism on the part of signal targets, our last predicted main effect is that a lack of conflict will result in more favorable reactions than the conflict conditions.

In addition to these main effects we also predict several interactions. Regarding signal strength and information, we predict that the signals will have the greatest impact on the participant’s responses when a participant has no information about whether the character is lying or telling the truth so that the coefficient for signal strength will be greater in the private information condition relative to the public information conditions. In instances where participants have information that the sister is likely cheating, we predict that the costlier signals will be more aversive than the less costly signals, in opposition to our prediction on the main effect of signal strength.

Due to the same reason underlying our prediction for the main effect of conflict, we also predict two interactions with conflict. The first is that the effect of signal strength will be larger in the conflict condition compared to no conflict condition. The second is that private information will have a greater negative impact on helping in the conflict condition than the no conflict condition.

Materials and methods

Results

Our total sample size was \(N = 1878\). After removal of participants who either did not complete the study or who failed our pre-registered attention checks, our sample size was reduced to \(N = 1631\). A further 184 participants were removed either because they shared an IP address with more than 5 other participants, and were therefore suspected of being a single person, or who finished the survey in less than 215 seconds (the shortest time we were able to complete the survey ourselves) and gave a short, perfunctory response to a free-response item. All such participants were removed prior to any further anayses of the data. Basic demographic information on the remaining \(N = 1447\) participants (Male = 700, Female = 735, unspecified = 12) is in Table X. For the number of participants from each US State, see Figure SX.

Time 1 manipulation checks

At time 1 (T1), participants were randomized into 1 of 2 conflict conditions (conflict vs. support), and 1 of 3 information conditions (cheating, private information, honest), for a total of 6 conditions. Our study was based on the expectation that participants would exhibit less support for the sister in the conflict vs. support conditions, and less support in the cheating vs. private information condition, and more support in the honest vs. private information condition. As we said in our preregistration:

We predict that information that the character in the vignette is honest will result in more favorable reactions than the no information condition, which should receive more favorable reactions than conditions where there is information that indicates cheating.

We also predict that the conflict conditions will result in less favorable reactions than the non-conflict condition.

In every condition, we predict that emotional responses will track instrumental responses.

To confirm that our manipulations had the desired effects, we analyzed our T1 outcome variables. All 11 T1 outcome variables regarding the sister were strongly positively correlated with each other, except for anger, which was negatively correlated, and sadness about the niece’s health problem, which was about the niece and not the sister (and was weakly positively correlated). Among the positively correlated sister variables, correlation coefficients ranged from 0.4 to 0.77, with a mean of 0.54. We therefore conducted a Principle Components Analysis (PCA) of our T1 vars. PC1 accounted for 53% of the variance, and all T1 variables loaded equally on this component except for sadness about the sister’s daughter’s health problem, which loaded primarily on PC2. See figures 1 and 2 (A Table showing the loadings for our PCA can be found in Appendix 2).

Figure 1. Loadings plot showing how each of the variables included in our T1 PCA load on the first two principle components.

Figure 2. Biplot of PC1 and PC2 at T1. The arrows represent the loadings of each variable on the first two prinicpal components. Each dot is one participant.

We then confirmed that our T1 manipulations had the predicted effect on these outcome variables by fitting a linear regression model of PC1 as a function of the conflict and information conditions. As predicted, PC1 was higher (i.e., more positive toward the sister) in the support vs. conflict condition, and PC1 was also higher in the honest vs. private vs. cheating conditions. See Figure 3 and Table 3.

Figure 3. Manipulation checks: the effects of conflict and information conditions on reactions to helping the sister.

Table 3. Anova of linear regression of PC1 as a function of the conflict and information conditions.

term	sumsq	df	statistic	p.value
conflict	608.5	1	128.1	0
p_info	1703.0	2	179.2	0
Residuals	6643.7	1398	NA	NA

We predicted the following interaction between conflict and information:

We also predict that in the private information and cheating vignettes there will be an interaction between the amount of information and levels of conflict such that private information will have a greater negative impact on helping in the conflict condition than the no conflict condition.

Contrary to our prediction, there was no significant interaction between conflict and information in their effects on PC1 at time 1 (\(p = 0.33\)).

Time 2 Signal Effects

In order to see if our signal manipulations resulted in the desired perceptions (I.E. the depressed manipulation resulted in perceptions of depression), we examined the percent of participants that checked each box of our periceved emotional state manipulaton check sorted by the signal they saw (Visualized in Figure 4). In doing so it appears that our signal manipulations were largely identified correctly, with \(85\)% and \(87\)% of participants reporting the sister as suicidial in the suicide attempt condition and the depression plus display of suicidial ideation respectively. Similarly, \(87\)% of participants in the depressed condition viewed the sister as depressed, and \(89\)% participants in the schizophrenic condition viewed the sister as mentally ill.

Figure 4. How participants in each signal condition rated the sister’s mental state. Participants could check any number of conditions. Colors represent the percent of participants that checked each condition. Red = low, Brown = middle, Yellow and White = high.

## quartz_off_screen 
##                 2

In addition to allowing us to confirm that the signals worked as designed, this question also allows for the examination of how the signals affected one’s perception of the sister’s current state and what baseline assumptions were being made in the control condition. Of particular interest is that those in the control condition often came in with the assumption that the sister was sad, scared, depressed and distressed despite not seeing a signal in the second portion of the vignette, with each of these option being checked by at least half of the participants.

To better understand which combinations of responses were most meaningufl, we also ran a PCA on all the variables that recorded if participants percieved the imagined sister to be in the emotional states listed above. PC1 seems to capture perceptions of sadness, fear, depression, and distress (Figure 5), while PC2 is heavily influenced by perceptions of mental illness (Figure 6). See appendix x for the loadings plot and biplot.

Figure 5. Boxplot showing the loadings on PC1 for each of the signals for our second PCA looking at perceptions of the sister’s emotional state.

Figure 6. Plot showing the proportion of participants who percieved their imagined sister to be mentally ill. Perceptions of mentall illness is plotted on the x-axis instead of PC2 due to both PC2 appearing to represent primariliy percpetions of mentall illness and because perceptions of mental illness are the more theoretically relvant variable.

Preregistered tests

Our predicted main effects of signals were:

We predict that stronger signals will result in higher levels of both emotional and instrumental social support towards the signaler. Specifically, we predict that the signals have the following rank order form low to high: no signal (control), verbal request, sad face, sad face with crying, depression, depression with display of suicidal ideation, and suicide attempt. We also predict that anger will have a positive impact on helping with a similar magnitude to the depression condition.

This prediction was not upheld. The Control condition (no signal) was ranked relatively high rather than the lowest, Suicide attempt was ranked relatively low rather than the highest, and Anger was consistently ranked lower than Depression. In partial support, however, Depression was consistently higher than Crying, which was consistently higher than Verbal request. See Figure 7.

Figure 7. The effect of the signal manipulations on changes in (A) the amount of money the participant is comfortable lending the sister at T2, controlling for the amount at T1 (dotted line indicates mean amount at T1), (B) the likelihood of lending the sister money, (C) participants’ perception of the sister’s need for money, and (D) the effect of the signal on PC1 of all the outcome variables at T2, controlling for t1. Effects are averaged across all time 1 conditions.

We also predicted:

Unlike the hypothesized signals of need, we predict that schizophrenia, which also comes with great fitness costs, will decrease the likelihood of help relative to the control since it indicates cognitive dysfunction.

This prediction was strongly supported: Schizophrenia had far and away the strongest negative impact on participants’ emotional attitudes towards, and instrumental support for the sister.

Our predicted interactions of signals with private information were:

In addition to these main effects, we also predict that the signals will have the greatest impact on the participant’s emotional state and likelihood to help when a participant has no information about whether the character is lying or telling the truth. Specifically, we expect the coefficient for signal strength to increase in the private information condition relative to the public information conditions.

We also predict that when the most severe signals, i.e., depression and suicidality, are paired with information that the character in the vignette is likely to be lying, responses will be more aversive than when the signal is weaker.

Like with the main effects of signal type, we fit linear models for our four outcome variables: (1) the amount of money the participant is comfortable lending the sister at T2, controlling for the amount at T1, (2) the likelihood of lending the sister money, (3) PC1 of all the outcome variables at T2 controlling for PC1 at T1, and (4) participants’ perception of the sister’s need for money. Our prediction was only supported when looking at our model with the amount of money participants were comfortable lending at T2 (controlling for T1) (\(p = 0.0045\). See Figure 8 for effects plot and Appendix 3 for the corresponding ANOVA tables.

Figure 8. Signal and information interactions. The effect of the signal, information interaction on changes in (A) the amount of money the participant is comfortable lending the sister at T2, controlling for the amount at T1 (dotted line indicates mean amount at T1), (B) the likelihood of lending the sister money, (C) PC1 of all the outcome variables at T2, and (D) participants’ perception of the sister’s need for money. Effects are averaged across all time 1 conditions.

Our predicted interactions of signals with conflict were:

We predict the coefficient for signal strength to increase in the conflict condition than in the no conflict condition.

We also fit linear models for the same outcome variables to examine test our hypothesized interaction. Unlike the predicted signal and information interactions, 3 of the 4 models showed evidence of interactions with the model for change in likelihood of lending T2 being marginally significant (\(p = 0.054\)). See Figure 9 for effects plot and Appendix 4 for the corresponding ANOVA tables.

Figure 9. Signal and conflict interactions. The effect of the signal, conflict interaction on changes in (A) the amount of money the participant is comfortable lending the sister at T2, controlling for the amount at T1 (dotted line indicates mean amount at T1), (B) the likelihood of lending the sister money, (C) PC1 of all the outcome variables at T2 controlling for T1, and (D) participants’ perception of the sister’s need for money. Effects are averaged across all time 1 conditions.

See Appendix 5 for a full overview of the full 2x3 interaction with change in perceived need at time 2 as the outcome variable.

Exploratory analyses

For our exploratory analyses, we restricted our attention to three signals: verbal request, crying, and depression, which we ranked in that order, and to two information conditions: private information and honest, which we ranked in that order. We then compared the causal effect of the signal on emotional and instrumental responses towards the sister.

To better understand where depression separated itself from the less costly signals we used a loess regression to visualize how PC1 T1 and PC1 T2 varied by signal in both the conflict and supportive manipulations (Figure 10). In doing so, it appears that the depression condition resulted in relatively more beneficial responses on PC1 T2 when participants saw the conflcit manipualtion and were intermediate on PC1 T1, with depression not separating itself from the other signals when participants were high or low on PC1 T1 to begin with or when the relationship with the sister was supportive.

Figure 10. PC1 at time 2 as a function of PC1 at time 1, by signal and conflict condition. Each dot is one participant. Lines fit by loess regression.

We also created an exploratory model based largely on theory but refined by observations of the data that suggested that much of the effect of the information manipulation was through its main effect rather than through its main effect plus the predicted interactions (Figure 11 and Table 4). Despite this being our favored exploratory model, it is unclear if private information would not interact much with signal type in real situations or if our conflict manipulations were simply more succesful than our information manipulations in activating the relevant mechanisms.

Figure 11. Effects plot showing the interaction between signal type and conflict in our favored exploratory model.

Table 4. Anova of interaction model depicted in Figure X.

term	sumsq	df	statistic	p.value
(Intercept)	51.3	1	10.7	0.0
signal	133.3	2	13.9	0.0
conflict	9.2	1	1.9	0.2
p_info	40.9	1	8.5	0.0
signal:conflict	36.1	2	3.8	0.0
Residuals	1652.1	344	NA	NA

Mediation model

As changes in perceived need were strongly correlated with our measures of instrumental support, we fit multiple mediation models to see how this was influenced by signal type using the ‘mediation’ package in R (Imai, Keele, & Tingley, 2010). See Figure 12.

Figure 12. Mediation model of causal effect of depression signal on change in likelihood of lending money, relative to verbal control.

Our first model compared the depression to the verbal reuqest condition. According to this analysis, the total effect of the depression signal was to increase the likelihood of lending money by 13.2 points, relative to a verbal request; 100% of this effect was mediated by the increased perception that the sister needed money. See Figure 13.

## quartz_off_screen 
##                 2

In the model comparing depression to crying, the direct effect of the depression signal was to decrease the likelihood of lending money by points relative to a verbal request, while the mediated effect increased the liklihood of lending money by xxxx points.

Lastly, in the model comparing depression to the control, the direct effect of the depression signal was to decrease the likelihood of lending money by points relative to a verbal request, while the mediated effect increased the liklihood of lending money by xxxx points.

Unlike the depression and verbal request model, the proportion of the relationship that was mediated by changes to perception of need was not reported for the follwoing models due to the changes in sign between the direct and mediated effects making such estimates meaningless (Hayes & Rockwood, 2017)

Discussion

Due to the theoretical and empirical similarities between the proposed signals of need, this study sought to test for the existence of important trends that transcend signal type. In particular, costlier signals were expected to result in more beneficial responses than less costly signals, with signal strength being predicted to matter most in times of private information or conflict. However, when the signals were viewed as a continuum based on costliness, the results of this study were mixed in relation to our hypotheses drawn from signaling theory with some signals fitting our predictions better than others.

Consistent with our predictions, both one’s reported likelihood of helping and the amount participants felt comfortable lending was greater in the no conflict condition compared to the conflict condition both before and after the signal was seen. As previously mentioned, this may be due to existing conflicts of interest increasing the amount of need the signaler would have to be in to be worth helping or the possibility that conflict might increase one’s suspicion of cheating in others. However, the finding that the presence of conflict did not increase the negative effect of private information on responses towards the signaler compared to the honest condition does not support the latter interpretation.

As with conflict, we found support for our predictions about the role of information, with private information resulting in less help than the public information of honest need condition but more than the public information of cheating condition. Interestingly, the private information condition resulted in outcomes closer to the honest condition than the cheating condition before participants saw the signal. However, it is unclear if this stems from the low costs of displaying generosity in this experiment (as no real investment is given) or if people tend to fall closer to believing need when lacking information in the condition we created. If the latter is the case, it may be driven by the benefits provided through kin selection which make investing in saving one’s niece well worth the lack of non-essential investment in one’s daughter, assuming the relevant mechanisms were activated correctly.

Despite both conflict and information having effects on how people responded to the proposed signals, it is clear that they do not account for the totality of the variance seen, which was substantial. Part of this may stem from personality differences, with variation in generosity, baseline trust levels, and willingness to lend money all being likely candidates . In addition to this, it also likely that differences in the success of the vignettes in eliciting the desired thoughts and emotions also mattered, with personality differences and past experience potentially interacting to influence this. Although this cannot be directly measured in this study, demographic variables, which are likely to be correlated with certain experiences, did not have an effect on helping outcomes pre- or post-signal.

Unlike with conflict and information, our predictions regarding the effects of the signal manipulations on helping were largely unmet when looking at the proposed signals as a whole. Of these predictions, only the schizophrenic break condition performed as expected compared to the other signal manipulations, consistent with our hypothesis that it indicates malfunction rather than a response to need. The reason for this overall lack of success is that the less costly signals performed similarly to the costlier signals across many conditions, with the exception being the suicide attempt condition, which was routinely outperformed by crying in all but the conflict and cheating condition.

Since our predicted rank order was not met, our results are likely best interpreted by focusing on the outcomes of each signal and through comparisons from signal to signal. Of all the proposed signals of need, depression tended to result in the most favorable responses emotionally and instrumentally. This coupled with the fact that it was largely not considered indicative of mental illness suggests that the depression symptoms presented did convincingly display need in a way that facilitated investment. Importantly, this still happened despite all the symptoms being presented in one-time events, something which likely biased our results against our predictions due to the fact that the extended length of its symptoms is what makes depression especially costly.

Although depression outperformed the other signals when averaged across conditions, it did not outperform the less costly signals by much, something which makes it reasonable to question if the extra costs involved were worth the increased benefits on average. Unfortunately, this cannot be addressed directly with the data we collected. However, it is important to note that looking at the responses collectively masks both the fact that depression separated itself from the less costly signals in instances of conflict and the fact that it still performed close to more appropriate signals when it was less than ideal. This suggests that in similar scenarios, most of the costs of an unnecessary attempt at signaling likely come through the costs of signaling rather than aversive responses, something which likely reduces the risk of depressive displays in times of extreme need.

Although not explicitly predicted as part of our preregistration and examined only in exploratory analyses, it appears that part of depression’s success in eliciting positive responses relative to the less costly signals came from its ability to increase the perception of honest need in others. This is seen in both the strong correlation between perceived need and our measures of instrumental support across signals and the finding that 98% of the effect of depression on helping relative to verbal request was through depressions effect on change perceived need. Although not predictied in our preregistration, this finding is consistent with costly signaling theory, which views the costs involved with a signal solely as a way to increase the perception of honest need in others

Despite depression’s relative success in eliciting positive responses relative to the other signals, it did not result in large increases in either our emotional are instrumental outcomes, nor did any other signal. Although we made no predictions to the contrary, this finding is surprising as the perceived likelihood of another’s need being honest (of which costly signals are expected to increase) is expected to play a large role in how one responds to requests for more support (Hagen, 2003). Among the reasons this happened, the most likely is that participants expected depression symptoms before the signal in response to the severity of the imagined sister’s need. Consistent with this, the most common combinations of responses to our question about how participants in the control perceived the sister’s emotional state included depression and the other emotions that loaded similarly on PC1, suggesting that depression symptoms here largely served to corroborate rather than enhance one’s appearance of need.

Compared to depression, the suicide attempt condition resulted in relatively negative responses across manipulations. Among the reasons for this that still allow for some suicide attempts to be the result of adaptations for bargaining, the most likely is that it was too costly for the situation. Although signal strength is often conceived so that costlier displays invariably increase the honesty of the signal (Zahavi, 1993), the believability of need is only one component in deciding if one should provide support and the size of such support. In this scenario, the fact that the niece was very young likely meant the probability of help mattering in the long-term relied heavily on the capability of the mother to provide care, whose display of suicide may have indicated this was in question. In line with this reasoning, those in the suicide condition often reported the mother as being mentally ill, something that was largely absent from those who read the depression condition.

This potential mismatch between signal strength and the severity of one’s situation may also explain why suicide attempts did not result in increased perceptions of need as predicted by its costs alone. One way this might happen is if suicide attempts appear manipulative in times of minor need or situations in which a signal is not needed. Although we did not ask if participants viewed the sister’s request to be manipulative directly, those in the suicide attempt condition perceived the request as more indicative of an attempt to get the money for other purposes than those who saw less costly signals. In line with the manipulation explanation, this pattern was reliably seen across combinations of the information and conflict manipulations except for the private information with conflict condition, when signal strength is expected to matter most. However, the fact that it was so closely correlated with our other outcome variables prevents any firm conclusions about whether this relationship was causal.

Whatever the underlying cause of the suicide attempt condition resulting in less help than the less costly signals of need, it appears clear the attempt itself caused it. This is due to the fact that while both the suicidal ideology and suicide attempt vignettes were built off of the depression vignette, the display of suicidal ideology condition resulted in responses much more similar to the depression condition than the suicide attempt condition in terms of perceived need and instrumental outcomes. However, when it came to perceptions of mental illness, the display of suicidal ideation condition resulted in responses much more similar to the suicide attempt condition than the depression condition, suggesting that important distinctions were being made within the mental illness category so that attempters were still treated differently.

Besides the suicide attempt condition resulting in less help than the other proposed signals, the biggest surprise was how well crying did in communicating need and eliciting support. Although the costs of crying are likely to be extremely small compared to depression (Hasson, 2009), crying resulted in similar levels of perceived need in most conditions, with depression mainly outperforming it in our monetary support measure. As previously mentioned, part of this may have to do with the scenarios we created resulting in situations in which costlier signals were not needed, something which may be especially likely in the honest and supportive manipulations.

It is also conceivable that crying may be an honest display for reasons outside of its costliness. One way this might happen would be if it is somehow linked to one’s underlying state in ways that could not be faked. Such a situation has been suggested to be the case with emotions and facial expressions , with certain expressions being difficult or impossible for most people to fake (Ekman, 2003). Of all the emotions tested, sadness was associated with the highest number of difficult to fake expressions, suggesting that crying’s co-occurrence with these expressions could result in honest displays. In addition to this, studies of infant crying have shown that cries differ in response to different inputs and that parental responses often track this, with pain cries developing more rapidly and to greater intensities than other types of crying (Zeifman, 2001). Assuming cries differ similarly in adults, it is possible that some types of crying can be assumed as honest and that participants may have used their personal experience with such cries to influence their responses when the signal was just imagined.

The final area where our main effects predictions were not borne out regarded the anger condition. As aggression and depression have been suggested to be alternative ways to accomplish similar bargaining outcomes (Hagen & Rosenström, 2016), we predicted anger to have a similar effect to depression on helping behavior. However, anger resulted in less help in both measures. Unfortunately, the fact that the signaler was always a woman means that this study is an insufficient test of this hypothesis as formidability is strongly confounded with sex (Sell, Tooby, & Cosmides, 2009). Therefore, the results of the anger condition may be driven by displays of anger being a less effective signal in women as they are unlikely to bargain as well through aggression. In addition to this, the likelihood that anger also serves as a way to downregulate one’s welfare tradeoff ratio with another further limits the ability of this study to evaluate anger as a signal of aggression as this was not controlled for (Sell et al., 2009).

Appendices

## quartz_off_screen 
##                 2

Appendix 5

Effects plot showing the effects of each manipulation on the change in the perceived need of the sister at time 2. The y-axis is ordered based on our predictions, with signals expected to result in more support as one moves from bottom to top.

Appendix 6

Loadings plot showing how each of the variables included in our sister’s emotions PCA load on the first three principle components.

Biplot of PC1 and PC2 for the sister’s percieved emotional states. The arrows represent the loadings of each variable on the first two prinicpal components. Each dot is one participant.

Appendix 7

How easy participants percieved imagining themselves in this situation based on manipulation seen. (A) signal by information. (B) signal by conflict. Higher values indicate greater ease (0 = Extremely difficult; 100 = Extremely easy).